Hi everyone,
I am trying to navigate the literature of how to process pupil data and have some questions about what to do with blinks. We are using SR Research EyeLink 1000 Plus with headrest. We are not utilizing any gaze positioning just looking at pupil size change during a cognitive task. We have a baseline of 250ms (-250ms to 0ms) and then stimulus presentation then 2000ms interest period (sampling at 1000Hz).
Since it is measured in arbitrary units we are doing a baseline correction to normalize the data to the baseline for each trial.
There seems to be mixed literature on interpolating blinks. Do it, not do it? Linear interpolation, cubic spline interpolation, etc.
I am a bit confused on:
1. Do we interpolate blinks during the baseline period?
2. How do we do time series data and normalize to baseline if there is fall out from blinks?
3. For the interpolation during each trial, which method is best?
When doing a cubic spline interpolation using R and it worked well for the dilation of the eye, but the same interpolation ran on the baseline of that trial did not look right.
No smoothing or filtering has yet been applied to this data
Hi Danielle,
Because, as you say, there are so many inconsistencies in the literature as to how pupil preprocessing should go, I like to follow the pipeline suggested by Kret & Sjak-Shie:Article Preprocessing pupil size data: Guidelines and code
They also published a code which you can use on your data.
Answering your questions briefly:
1. The blinks should be interpolated on continuos signal, not on a small window (like a baseline). However, if due to a blink you are missing too much data in the baseline, you may need to reject this trial, as you have no "real" signal to calculate the baseline from.
2. I'm not sure I understand what you mean "fall out form blinks" - is it maybe that due to the blink you have data loss in the time window of the baseline? If so, please see the point above.
3. This is a matter of discussion, of course, but again, I suggest you take a look at Kret & Sjak-Shie's paper. For a more elaborate discussion, take a look at (many) papers by Mathôt, maybe especially Article Safe and sensible preprocessing and baseline correction of p...
About your last comment, I have a feeling that you run your preprocessing separately for the time windows of the stimuli and of the baselines - is that correct? Remember that you deal with a continuous signal, so things like interpolation should not be used on cut-out parts of that signal. If I misunderstood, sorry about it. Either way - the papers mentioned above should help a lot!
Good luck,
Lena